Individual-Based Exposure Assessment to Traffic-Related Air Pollution: Comparing Land Use Regression and Dispersion Modelling in Multiple European Cities

Abstract

Individual-Based Exposure Assessment to Traffic-Related Air Pollution: Comparing Land Use Regression and Dispersion Modelling in Multiple European CitiesAbstract Number:1985 Kees de Hoogh*, Michal Korek, Jaakko Kukkonen, Menno Keuken, Gerard Hoek, Tom Bellander Kees de Hoogh* MRC-PHE Centre for Environment and Health, Department of Epidemiology and Biostatistics, Imperial College London, United Kingdom, E-mail Address: [email protected] Search for more papers by this author , Michal Korek Institute of Environmental Medicine, Karolinska Institutet, Sweden Search for more papers by this author , Jaakko Kukkonen Finnish Meteorological Institute, Air Quality, Finland Search for more papers by this author , Menno Keuken Netherlands Applied Research Organization, Netherlands Search for more papers by this author , Gerard Hoek Institute for Risk Assessment Science, Utrecht University, Netherlands Search for more papers by this author , and Tom Bellander Institute of Environmental Medicine, Karolinska Institutet, 6. Centre for Occupational and Environmental Medicine, Stockholm County Council, Sweden Search for more papers by this author AbstractFew comparisons have been made between the use of land-use regression (LUR) and dispersion (DISP) models to predict small-scale variation in air pollution levels for epidemiological studies. Within the European Study of Cohorts for Air Pollution Effects (ESCAPE) we have made direct comparisons between LUR and dispersion models for NO2 and PM10Methods. The ESCAPE study developed LUR models for air pollution levels based on a uniform monitoring campaign. In several ESCAPE study areas we were also able to apply DISP models; most of the study areas used Gaussian plume models for localized sources. Some used street-canyon models and some Eulerian or similar models for regional and urban background. We compared the two methods at the cohort’s participants residence addresses. Additionally we compared the DISP model estimates with measured concentrations at the ESCAPE monitoring sites (that were used in developing the local LUR prediction models).Results. LUR and DISP model predictions for NO2 showed an average R2 of 0.52 for 111,476 addresses over 12 study areas. The median and range of predicted NO2 concentrations between LUR (26.33; 48.63 µg/m3) and DISP models (28.5; 45.33 µg/m3) were also similar. For PM10 the average R2 was 0.23 when comparing the 2 methods for 68,096 addresses over 6 study areas, with the median and range of the predicted PM10 concentrations for LUR (24.4; 15.8 µg/m3) and for DISP models (23.1; 18.2 µg/m3). The correlation between DISP model estimates and measurements showed R2’s ranging from 0.15 to 0.74 for NO2, (median R2 = 0.56, n = 12) and from 0.13 to 0.77 for PM10 (median R2 = 0.39, n = 6).Conclusions. LUR and DISP estimates correlated better for NO2 than for PM10. DISP models predict a moderate to large portion of the variation measured across the ESCAPE measurement sites, again more so for NO2 than for PM10.